Microsoft Wants Companies to Verify What AI Agents Do Before Handing Them More Control
Microsoft announced an AI Transparency Initiative at Ignite on November 17, seeking open standards for independently checking what enterprise agents do.
Microsoft is asking businesses to give AI agents more responsibility while admitting that today's controls cannot fully explain what those agents do. At Ignite on November 17, it announced an AI Transparency Initiative with more than 40 organizations to develop common standards for checking their behavior independently.
The problem is bigger than keeping a transcript. Agents can use company data, credentials and tools to act without a person approving every step. Microsoft says existing controls cannot reliably reconstruct everything between a request and its outcome.
An audit trail beyond the model's account
Microsoft plans to expand its work with NIST's agent standards initiative and bring the resulting approach into Microsoft AI, Foundry and Copilot.
The stated goal is evidence that security and compliance teams can use across providers, including when developers switch models. That is a standards effort and a product direction, rather than proof that the visibility gap has already closed.
Containment is moving into Windows
A separate Windows announcement at Ignite described new containment capabilities in Microsoft Execution Containers, with OpenAI and Meta already building on them. Microsoft also said Windows administrator protection and its endpoint security platform are now generally available.
The announcements address different parts of the same problem. Containment limits where an agent can operate; independent evidence helps establish what happened inside those limits. Neither an activity log nor a container, by itself, answers every question about an agent's decisions.
Microsoft's transparency announcement leaves accountability with the organization deploying the agent. Giving software permission to act does not transfer responsibility when the action goes wrong.
Some offers on this page may be paid placements or contain affiliate links.
Cyberpresso: daily cyber & AI brief
Free daily newsletter, read in 5 minutes.
Subscribe free